Tejal Patwardhan
@tejalpatwardhan
thinking hard about hard evals // @openai
all aligned models look the same but every misaligned model is misaligned in its own way
It was a *character-building* privilege to post-train GPT 4.5
GPT-4.5 is ready! good news: it is the first model that feels like talking to a thoughtful person to me. i have had several moments where i've sat back in my chair and been astonished at getting actually good advice from an AI. bad news: it is a giant, expensive model. we…
worth looking at the bio results.
OpenAI and Anthropic *both* warn there's a sig. chance that their next models might hit ChemBio risk thresholds -- and are investing in safeguards to prepare. Kudos to OpenAI for consistently publishing these eval results, and great to see Anthropic now sharing a lot more too.
deep research system card is out! We included new SWE-Lancer results as part of our Preparedness evaluations. Deep research (with browsing) achieves SOTA on SWE-Lancer Diamond, earning $259K / $500K and solving 46% of IC SWE and 51% of SWE Manager tasks.
We're also sharing the system card, detailing how we built deep research, assessed its capabilities and risks, and improved safety. openai.com/index/deep-res…